Repository file with inexplicable branch revisions

[Doyle, Jim]

Hi, 

One of our development groups started up a CVS repository recently and
has had some troubles with it, where it looks like the repository has
gotten corrupted.  I've narrowed it to a single repository file, which has
headers that look like:

head    1.1.1;
access;
symbols
        production:1.1.1
        ISO:1.1.1;
locks; strict;
comment @# @;


1.1.1
date    2001.11.05.21.08.47;    author rslate;  state Exp;
branches;
next    ;


desc
@@


1.1.1
log
@Initial entry


There are no obvious glitches as a result of direct edits,
like missing lines or added characters, but the revision history is
extremely odd.
There is only one revision, called "1.1.1"!  The file was originally added
to the repository via an import command like "cvs import ncpc ISO
production".
Normally the revisions would be "1.1.1.1" and "1.1", with the vendor branch
tag
on "1.1.1".  The effect on the CVS client/server is that the module cannot
be checked out, with the error "bad branch tag in checkout", which sounds
only reasonable given the strange revisions.

It appears that the strange revision state must have been caused by a
sequence
of legitimate CVS commands, but I've been trying to reproduce it by 
moving tags, deleting tags, outdating revisions, and committing with
explicit
revisions, and I haven't gotten anywhere.  The repository is under CVS
1.11.1p1 running 
as pserver on a Unix box.  Can anyone think of a sequence of
CVS commands that would result in this case, with the only revision having a
branch number like "1.1.1"?

At this point the best we can do is remove the repository file from the
server and 
re-add it (since it's been corrupted since the very day it was added).  But
I want
to find out what operations to warn people against, or what I can do to
protect
the repository from these operations.

Thanks in advance,
Jim Doyle
address@hidden


 <<ncpc.sps,v>>

ncpc.sps,v
Description: Binary data